Data from McKinsey’s Latest AI Survey

Last week, McKinsey published a global survey of AI adoption among 1,993 respondents, 38% of whom work for organizations worth $1B and above. The largest companies are moving beyond simply testing AI and are busy implementing agentic workflows in key areas. But it’s still the early days and few companies say that AI agents are actually performing at scale for them. 

Almost 9/10 respondents say their companies use AI somewhere. Only a minority see profit moving at the enterprise level. 88% report AI use in at least one function while just 39% report any EBIT impact. 

Though there is a small “high performer” group, about 6% of respondents, that reports ≥5% EBIT from AI and shows consistent behaviors that others do not. They redesign workflows, scale faster, and put real budget and leadership behind the work.

What surprised me most was the mix of progress and risk. 51% of AI-using organizations say they have already experienced at least one negative consequence. Nearly a third call out AI inaccuracy and explainability. That explains why regulated teams and fast-moving commercial groups keep pilots narrow until controls catch up. It also explains why a small group pulls ahead. Those high performers adopt more risk controls early and move on.

Agents are real but still thin in any one department. McKinsey finds 62% of businesses are experimenting with agents and 23% scaling at least one agent somewhere. In a given function, fewer than 10% have agents at scale.

The report also finds that scale still favors size. About half of companies with +$5B in revenue report they have reached the scaling phase. Only 29% below $100M say the same. Budget is part of it. More than one third of high performers commit +20% of the digital budget to AI and roughly three quarters say they are scaling or have scaled. The other part is leadership. McKinsey links senior ownership to impact. Teams that see value have leaders who set growth goals, not only cost goals, and who push workflow changes, not just tool rollouts. 

So what actually drives impact? Value shows up first in a few measurable jobs with clear owners. Software engineering and IT report cost benefits early, marketing and sales, strategy and finance, and product development report revenue effects. 

Then the playbook spreads –  embedding tools in routines, tracking a small set of KPIs, and redesigning the steps that people follow, not only the app they click. High performers repeat these moves and move faster on agents. They also feel more pain when things go wrong because they are using AI in more places – that is the trade. More value and more exposure handled with better guardrails.